Model, analysis, and improvements for inter-vehiclecommunication using one-hop periodic broadcasting based onthe 802.11p protocolCitation for published version (APA):Batsuuri, T., Bril, R. J., & Lukkien, J. J. (2011). Model, analysis, and improvements for inter-vehiclecommunication using one-hop periodic broadcasting based on the 802.11p protocol. (Computer science reports;Vol. 1109). Technische Universiteit Eindhoven.

Document status and date:Published: 01/01/2011

Document Version:Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can beimportant differences between the submitted version and the official published version of record. Peopleinterested in the research are advised to contact the author for the final version of the publication, or visit theDOI to the publisher's website.• The final author version and the galley proof are versions of the publication after peer review.• The final published version features the final layout of the paper including the volume, issue and pagenumbers.Link to publication

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, pleasefollow below link for the End User Agreement:www.tue.nl/taverne

Take down policyIf you believe that this document breaches copyright please contact us at:openaccess@tue.nlproviding details and we will investigate your claim.

Download date: 07. Jan. 2021

Model, analysis, and improvements for inter-vehicle

communication using one-hop periodic broadcasting based on the

802.11p protocol

Tseesuren Batsuuri, Reinder J. Bril, and Johan Lukkien

Department of Mathematics and Computer ScienceEindhoven University of Technology

Eindhoven, The Netherlands

Abstract

Many future vehicle safety applications will rely on one-hop Periodic Broadcast Communi-cation (oPBC). The key technology for supporting this communication system is the newstandard IEEE 802.11p which employs the Carrier Sense Multiple Access/Collision Avoid-ance (CSMA/CA) mechanism to resolve channel access competition. In this work, we firstaim at understanding the behavior of such oPBC under varying load conditions by consid-ering three important quality aspects of vehicle safety applications: reliability, fairness, anddelay. Second, we investigate possible improvements of these quality aspects. We start witha clear mathematical model which gives the foundation for making an accurate simulationmodel as well as for defining new appropriate metrics to judge the aforementioned qualityaspects. We evaluate oPBC with a strictly periodic broadcasting scheme, i.e., each vehiclebroadcasts messages in a strictly periodic manner. The evaluation reveals that the hiddenterminal, or Hidden Node (HN), problem is the main cause of various quality degradationsespecially when the network is unsaturated. To be more specific, the HN problem reducesthe message reception ratio (i.e., reliability degradation) and causes unfair message receptionratios for vehicles (i.e., fairness degradation). Moreover, it causes long lasting consecutivemessage losses (i.e., delay degradation) between vehicles while they are encountering eachother, i.e., entering their Communication Ranges (CRs). In some serious cases, a certainvehicle could not successfully deliver any of its messages to a particularly destination vehiclethroughout an entire encounter interval of these two vehicles. We propose three simple buteffective broadcasting schemes to alleviate the impact of the HN problem. Though these solu-tions do not affect the message reception ratio (i.e., reliability) of the entire network, they doimprove the fairness and delay aspects. These solutions are fully compatible with the IEEE802.11p standard, i.e., they are application-level solutions and can be easily introduced inpractice.

1 Introduction

A rapid progress in mobile and wireless technologies in the last decade enables a wide spectrumof applications in the Intelligent Transportation System (ITS) domain targeting vehicle safety,transportation efficiency, and driver comfort. In recent years, many industry/government con-sortiums are formed around the world to carry out projects to investigate such applications: theVehicle Safety Communications consortium in the US [1], the Car2Car Communication consor-tium in Europe [2], and the Internet ITS consortium in Japan [3]. As a result of these efforts,

1

many interesting vehicle safety application scenarios are identified and their communication re-quirements are carefully examined. It is now becoming clear that most of these applicationswill rely on broadcast communication that comes in two flavors: event-driven and time-driven.In the event-driven (or emergency) case, a vehicle starts broadcasting a safety message for acertain duration periodically when a hazardous situation is detected and, hence, these messagesare not sent in a normal situation. In the time-driven case, each vehicle continuously performsone-hop periodic broadcast communication (oPBC) to pro-actively deliver a beacon messagewith its status information (e.g., position, speed) to the neighboring vehicles. The key idea ofsuch oPBC is to make each vehicle aware of its vicinity such that future vehicle safety applica-tions running on the vehicle will leverage this information to detect any hazardous situation ina timely manner. A lane change advisor and a forward collision warning application [1] are twotypical examples that rely on this oPBC. These applications require a frequency of 10 messagesper second with a maximum no message interval (or a tolerance time window) of [0.3sec,1.0sec][1, 4, 5]. In addition, these applications pose a strict fairness requirement on oPBC [6, 7], whereeach vehicle should have equal opportunity for using the shared channel. In this type of system,message loss is unavoidable (we explain the causes below); however, it must not be the case thatone or a few vehicles take all the loss, because this would result in these vehicles becoming adanger to their surrounding vehicles.

In this work, we focus on this oPBC from the vehicle safety application perspective. Particu-larly, we are interested in oPBC which is addressed in the IEEE 802.11p [8], IEEE 1609 standards[9], [10], [11], [12], and Society of Automotive Engineer (SAE) J2735 [13]. The 802.11p standardhas been designed specifically for inter-vehicle communication. Besides the regular support forhigher-layer protocols like IP, the 802.11p Medium Access Control (MAC) supports a short mes-sage protocol called WSMP (WAVE Short Message Protocol, IEEE 1609, where WAVE standsfor Wireless Access in Vehicular Environments). Among other uses, this WSMP protocol to-gether with the SAE J2735 addresses the transmission of Basic Security Messages (BSM) alsoknown as beacon messages that are used by a vehicle to inform other vehicles about its statusand condition. The BSM (in the rest of paper, we simply call it message) is sent periodically,in broadcast mode, with a typical frequency of 10Hz.

In general, the members of the 802.11 family, where the 802.11p is one of the newest members,of wireless standards support two communication modes: a managed mode called Point Coordi-nation Function (PCF) where a base station manages access to the channel and an ad-hoc modecalled Distributed Coordination Function (DCF) where stations collaborate to manage channelaccess [14]. In DCF, stations employ the Carrier Sense Multiple Access/Collision Avoidance(CSMA/CA) mechanism to resolve channel access competition. For point-to-point communi-cation, stations repeatedly perform channel sensing followed by a random Back-off (Bf) periodselected from an increasing Contention Window (CW). Bf is used to reduce the probability ofa contention problem which occurs when two or more stations that exist in each other’s Com-munication Range (CR) incidentally happen to start transmission at the same time causingcollisions. In addition, Request To Send/Clear To Send (RTS/CTS) signaling is used to resolvethe hidden terminal, or Hidden Node (HN), problem which occurs when two stations that areoutside each other’s CR have overlapping transmissions in time interfering with their commonneighbors in the intersection of the CRs. On top of this, a MAC level acknowledgement canbe used to resolve the remaining message losses. An initial channel access delay, namely Arbi-tration Inter Frame Space (AIFS), allows discriminating among several priority classes. Whenstations broadcast messages rather than sending them point-to-point the situation in DCF isquite different. First, CW from which a Bf period is drawn is fixed, and Bf is at most done once[14]. Second, RTS/CTS signaling and MAC layer acknowledgement do not work since there is

2

mj

mi

mi

i mi

pji

(c) HN problem(b) Contention problem(a) CSMA/CA

i

j

p

pj

AIFS

mi mi

mp

pji

mp

mj

mi Sending message of i mi Receiving message of i

Collision

Ready to transmit (RTT) Access Deferral (AD)

Remaining BfBackoff (Bf)

mj t

Figure 1: the 802.11p MAC CSMA/CA in broadcast mode. (a) shows the contention protocol thatserializes three stations. A station, which wants to transmit, first must check if the channel isfree for a duration of Arbitration Inter Frame Space (AIFS). In case the channel is found activeduring this duration, the intended transmission is deferred and a Bf is performed. (b) shows amessage collision due to a contention problem which happens if incidentally, two stations starttransmission at the same time (mainly because their Bf periods expire at the same time), i.e., aNeighboring Node (NN) collision. (c) shows a message collision due to the Hidden Node (HN)problem, i.e., a HN collision: two stations that cannot sense each other’s transmission maycause message collisions in the intersection of their Communication Ranges (CRs).

no particular destination for a message. As a result, when all stations use broadcast-based com-munication, the collision problems, i.e., the contention and the HN problems increase. Figure 1gives an overview of the 802.11p communication behavior in broadcast mode and illustrates thecollision problems.

The purpose of our research is to understand the behavior of this oPBC based on the 802.11pDCF. We want to understand message losses due to the contention and HN problems undervarying load conditions. In particular we want to understand oPBC by considering three qualityaspects which are important for vehicle safety applications: reliability (i.e., successful messagereception ratio), fairness (i.e., distribution of successful message reception ratio over vehicles)and delay (i.e., no message interval that is the longest interval in which a vehicle that is in theCR of another vehicle doesn’t receive a message from that latter vehicle.). In addition, we wantto investigate possible improvements.

Our first step is to develop a mathematical model of the 802.11p behavior under oPBC. Thisserves three purposes. First, it makes the discussion unambiguous. Second, it gives the foun-dation for a simulation model and third, it allows us to develop new relevant metrics (criteria)to judge communication quality for the above aspects. Standard performance metrics like asuccessful message reception ratio of the entire network or an average end-to-end delay fall shortin this environment. Our second step is to simulate according to this model and to determinethe values of the given metrics. This gives us insight in shortcomings of this oPBC and givesdirections for improvements.

The contributions of this work are fourfold. First, we give a novel timing model of oPBCunder the 802.11p DCF. Second, we define several new metrics for judging the quality of oPBC.Third, we simulate oPBC under several different circumstances showing problems and short-

3

comings. Namely, the simulation reveals that the HN problem is the main cause of variousquality degradations when the network is unsaturated. It reduces the message reception ratio ofthe entire network (i.e., reliability degradation) and causes unfair message reception ratios forvehicles (i.e., fairness degradation). Moreover, it causes long lasting consecutive message losses(i.e., delay degradation) between vehicles while they are encountering each other, i.e., enteringtheir CRs. In some serious cases, a certain vehicle could not successfully deliver any of itsmessages to a particularly destination vehicle throughout an entire encounter interval of thesetwo vehicles. Fourth, we identify application-level improvements. We propose three simple buteffective broadcasting schemes to alleviate the impact of the HN problem. These solutions arefully compatible with the IEEE 802.11p standard, i.e., they are application-level solutions andcan be easily introduced in practice.

The remainder of the paper is organized as follows. Section 2 introduces the mathematicalmodel of oPBC. Section 3 presents an evaluation study of oPBC by means of simulations whichare carried out according to the model and Section 4 presents our solutions for the collisionproblems. Section 5 covers recent studies most relevant to our work. Finally, Section 6 gives asummary of the main results and presents an outlook on future work. In addition, AppendixA gives a detailed overview of CSMA/CA in broadcast mode and Appendix B gives a shortoverview of a signal reception model which is chosen for our simulation model.

2 Models of oPBC

In this section, we introduce our model of oPBC. Based on the model, we show how the ap-propriate metrics are defined for evaluating the three quality aspects of oPBC (i.e., reliability,fairness, and delay).

2.1 Two models

In our analysis of communicating vehicles we encounter two aspects: a simulation of the move-ment of the vehicles and a simulation of the behavior of the wireless communication as a functionof the position of the vehicles. Thus we have the traffic model which yields the position of ve-hicles as a function of time and the communication model that describes the communicationevents between vehicles as a function of time and vehicle location. Hence, the communicationmodel depends on the traffic model (but not the other way around).

The interface between the two models is formed by the location of the vehicles. Togetherwith the radio channel model this yields the neighborhood structure viz., a set of vehicles thateach vehicle can transmit to or receive from at any point in time. The traffic model can be veryadvanced, even to the extent that life traces are simulated [15, 16]. In this work we are notconcerned, however, with the traffic model. For our simulations we stick to a simple highwaymodel, represented as a stretch of several kilometers with three lanes per direction and periodicboundary conditions (which makes it, in fact, a loop). Speeds per lane are assumed to be fixed.In simulations the main concern of the traffic model is to simulate with a small enough timestep to have a realistic and sufficiently accurate description for the communication model. Themotivation for this restriction is that we want to study just the communication model undervarying load conditions.

4

2.2 The communication model

The communication model consists of two parts: First, we describe the communication of the802.11p standard and the radio channel model that generates the events. Second we modeltiming and events in the system consisting of communicating vehicles to define the concepts ofinterest.

2.2.1 The 802.11p communication and the radio channel model

We restrict ourself to describing the broadcast mode of the 802.11p MAC. In Appendix A,Figure 21 gives an extended flowchart of the broadcast procedure taken from [17] and Figure 22describes the corresponding state machine diagram. In our simulation model, every vehicle isimplemented according to this state machine. Besides, we take a Signal to Interference plusNoise Ratio (SINR) based signal reception model of the updated NS-2 implementation of the802.11p [18] and a brief description of this model is given in Appendix B. In addition, we choosethe Two-Ray Ground (TRG) signal propagation model in order to study solely the effect ofmessage collisions. The main configuration parameters of the 802.11p and the TRG model arechosen as in Table 1.

Table 1: The 802.11p parameter settings

Parameters Values

Date rate 6MbpsA slot duration 13µs

AIFS 6 slotsCW size 7 slots

Th (Preamble length) 40µsAntenna gain 0dB

Antenna height 1.5mNoise floor(nF) -99dBm

Power Sense threshold(PsTh) –92dBmCarrier sense threshold (CsTh) -85dBm

SINR threshold (SrTh) 8dB

2.2.2 The timing model

We assume a set V of N vehicles v1, v2, ...vN periodically broadcasting messages. The behaviorof the system is described as a series of events happening at certain times. As a conventionwe use a superscript to denote a kth occurrence or instance. For example, e(k) denotes the kth

occurrence of an event e and m(k)i denotes the kth message of vi. In addition, we often do not

name the event but only the time of occurrence using a similar notation, as explained next.

The activation time a(k)i is the time at which vi becomes ready to broadcast m

(k)i . The

start time s(k)i and finish time f

(k)i are the times at which vi actually starts and finishes the

transmission of message m(k)i , respectively. Note, from a receiver vehicle’s perspective, the start

time and the finish time at which the vehicle starts and finishes receiving the message m(k)i are

s(k)i +δ and f

(k)i +δ, respectively. δ is an air propagation delay that is relatively small1, therefore

1δ � 1µs [19, 20]

5

we neglect this in our model.

The transmission interval tI(k)i of message m

(k)i is defined as

tI(k)i

def= [s

(k)i , f

(k)i ) . (1)

We require that

a(k)i < s

(k)i ≤ f (k)i ≤ a(k+1)

i (2)

holds. Message transmission is assumed to be periodic. If a message is not sent at all or isdelayed such that the remaining part of the interval is not enough for successful completionwe say that the message is dropped. This may mean a partial message transmission or, in the

extreme case, no transmission at all (s(k)i = f

(k)i ). In both cases, we define f

(k)i = a

(k+1)i and we

take that as the condition of message dropping.Moreover, we define transmission power Pti(t) of vehicle vi and its reception power at vehicle

vj as Prij(t) and cumulative reception power cPrj(t) of vehicle vj at time t. Note, we alwaysassume that i 6=j holds whenever we talk about two vehicles vi and vj . We require that Pti(t) > 0

holds during tI(k)i and its value is determined by the application. Prij(t) is determined by a

given signal propagation model, by Pti(t) and by the distance between sender and receiver attime t. cPrj(t) is determined by all receiving signal strengths at vj at time t plus a noise floor,nF, as follows

cPrj(t) = nF +∑vi

{Prij(t)|Prij(t) ≥ PsTh}, (3)

where PsTh is a Power Sense threshold of the receiver. Given these notions, we define theneighborhood of a vehicle. At any time t, each vehicle vi has a target neighbor set of othervehicles, Nbi(t), where vj ∈ Nbi(t) means that vj is in the CR of vi at time t. It is defined asfollows

vj ∈ Nbi(t)def=

Prij(t)

nF≥ SrTh, (4)

where SrTh is a SINR threshold for receiving the message successfully. Note, CR is the receptionrange, the places where the message could be received disregarding interference of other stations.A necessary condition for receiving a message is that the receiving vehicle must be in the CRof the sending vehicle for the duration of the message transmission. A sufficient condition fora message reception is that the receiving signal power must be equal to or greater than SrThwith respect to the cumulative power of all other signals for the entire duration of the messagetransmission. This is defined as follows

∀t : t ∈ tI(k)i ∧ Prij(t)

(cPrj(t)− Prij(t))≥ SrTh. (5)

We extend the concept of a neighborhood to intervals by

↓Nbi(I) =⋂t∈I

Nbi(t) . (6)

This interval represents all vehicles that have been in the CR of vehicle vi during the entireinterval I. Changes of neighbor sets are represented by enter and leave events. Entering time

e(k)ji is the time at which vj enters the CR of vi for the kth time while leaving time l

(k)ji is the

time at which vj leaves the CR of vi for the kth time. The kth encounter interval eI(k)ij of vj with

vi is defined as

eI(k)ij

def= [e

(k)ji , l

(k)ji ) . (7)

6

eji(z)

|eIij(z)|

lji(z)

ai(k)

si(k) fi

(k)

|tIi(k)|

mi(k+1)mi

(k-1)... ...

AIFS(s) + (AD(s)) + (Bf)

mi(k)

mi(k)

Figure 2: Vehicle vj enters the CR of vi at e(z)ji and leaves it at l

(z)ji . During this encounter

interval, vj receives a sequence of messages from vi. A vehicle vi becomes ready to broadcast

its kth message at a(k)i but it actually starts the transmission at s

(k)i and finishes at f

(k)i . The

distance between a(k)i and s

(k)i depends on the channel condition. In the best case, it can be only

an AIFS. In the worst case, it can be multiple AIFSs + multiple ADs(Access Deferrals) + Bf.

During eI(k)ij we say that there is a link from i to j and we call that the kth such link. Figure 2

gives a schematic of these notions.

Message loss The most important concern is whether messages are actually received by

vehicles that could receive them. Considering message m(k)i there are three reasons why another

vehicle vj might not receive it.

• (OOR) Out Of Range. In order for a vehicle vj to receive m(k)i it must be in the neighbor-

hood of vi for the duration of the transmission. When vj 6∈↓Nbi(tI(k)i ), vj does not receive

m(k)i .

• (MD) Message Dropping. This happens, as described above, if the back-off interval be-comes so long that the message transfer time does not fit in the remaining part of theperiod. In our model this is equivalent to

f(k)i = a

(k+1)i . (8)

No vehicle will receive message m(k)i .

• (MC) Message Collision. The message is transmitted but not received by vj since othervehicles may transmit at the same time to vj and their interferences are strong enough tocorrupt the receiving message of vi. This is defined as follows

∃t : t ∈ tI(k)i ∧ Prij(t)

(cPrj(t)− Prij(t))< SrTh. (9)

7

Given these reasons for loss we define the transmission condition of message m(k)i and, accord-

ingly, the reception condition of m(k)i by a vehicle vj as follows

Tc(k)i =

{MD if (8)XMT otherwise

(10)

Rc(k)ij =

OOR if vj 6∈↓Nbi(tI

(k)i )

MC if vj ∈↓Nbi(tI(k)i ) ∧ (9)

Tc(k)i otherwise.

(11)

If Tc(k)i = XMT, message m

(k)i is broadcast successfully. If Rc

(k)ij = XMT, the message is received

by vehicle vj at time f(k)i successfully.

Metrics We define the most appropriate metrics that can judge the communication qualityin the three most important aspects of the safety applications: reliability, fairness and delay.

For the reliability aspect, we use the fraction of successfully delivered messages (SMR, suc-cessful message ratio). This concept can be refined to links between vehicles and to individualmessages. To start we define the number of received messages from vi by vj in a given interval,as well as the number of times that such message could have been received. The ratio is thesuccessful message ratio in that interval.

Rsij(I) = |{k | tI(k)i ⊆ I ∧ Rc(k)ij = XMT}| (12)

Nsij(I) = |{k | tI(k)i ⊆ I ∧ Tc(k)i = XMT ∧ vj ∈↓Nbi(tI

(k)i )}| (13)

SMRij(I) =

{Rsij(I)

Nsij(I)if Nsij(I) > 0

0 if Nsij(I) = 0(14)

Generalizing this by summing over the receiving vehicles gives the successful message ratio of viin an interval.

SMRi(I) =

∑

vjRsij(I)∑

vjNsij(I)

if∑

vjNsij(I) > 0

0 if∑

vjNsij(I) = 0

(15)

As a special case, SMRi(tI(k)i ) is the SMR of m

(k)i . Again, generalizing by summing over the

sending vehicles we obtain the SMR of the entire network during that interval.

SMR(I) =

∑

vi,vjRsij(I)∑

vi,vjNsij(I)

if∑

vi,vjNsij(I) > 0

0 if∑

vi,vjNsij(I) = 0

(16)

At the network level an interesting question is: how does SMR([0, T )), where T represents atime of consideration, behave as a function of vehicle density?

From the fairness perspective, the behavior of individual vehicles is more important thanthe average. This is why we also analyze SMRi to see whether losses are distributed evenly (orfairly) over the vehicles. The cumulative distribution function shows this; a fair distributionwould give a transition from 0 to 1 within a short interval.

cdfSMR(I, x) =|{vj | SMRj(I) ≤ x}|

N, for 0 ≤ x ≤ 1 (17)

8

In addition, plotting SMRi as a function of time gives insight in the visibility of vi for othervehicles.

Finally, from the delay perspective, an important further question is how losses of a particularvehicle are distributed in time and across vehicles: do losses happen in sequences and do theyaffect the same links? To that end we define the concept of a “No Message Interval” betweentwo vehicles during a given interval I which is the length of the longest subinterval of I withouta successful message transmission. In addition, the “First Delay” is the length of the longestinitial subinterval and represents a delay in discovery in case we apply it to an encounter interval.

NoMij(I) = sup {|J | | J ⊆ I ∧ Rsij(J) = 0} (18)

FDij([a, b)) = sup {x | [a, a+ x) ⊆ [a, b) ∧ Rsij([a, a+ x)) = 0} (19)

In our analysis we look at genuine NoM and FD, viz., those that correspond to encounterintervals. These are examined as a function of their length and plotted as a density (histogram)or as a cumulative distribution.

3 Evaluation of oPBC under CSMA/CA coordination

In this section, we evaluate oPBC using the newly defined metrics. We implemented a simulatoraccording to the model and verified its correctness against the updated NS-2 implementation of802.11p [18]. The following subsections describe the simulation setup, results, and analysis.

3.1 Simulation Set Up

For the purpose of this evaluation, two different scenarios are simulated. In the first scenario(single domain (SD)), vehicles are deployed at fixed locations within a single CR viz., all vehiclescan receive each others messages. This scenario allows us to study the collisions caused only bythe contention problem, i.e., NN collisions since there are no HNs. In the second scenario (multidomain (MD)), vehicles are deployed on a 3km long highway with three lanes per direction. Thisscenario allows us to study both HN and NN collisions. By having these two scenarios, we cancompare the impact of these two types of collisions. The vehicles at the three lanes have fixedvelocities of 20, 30, and 40 m/s respectively. In both scenarios, different inter-vehicle spacingsare used in order to create different Vehicle Densities (VD). We assume a single channel, a fixedbroadcasting period and initially, a random phasing within this period as

a(k)i

def= φi + kTi. (20)

Thus, each vi has a broadcasting period Ti ∈ R+ and an initial broadcasting phase φi ∈ R+,where φi is uniformly selected from an interval of [0, Ti). Moreover, we assume the same signal

Table 2: Simulation settings

Parameters Values

Message size 555 bytesCR 300m

Broadcasting Period (T) 0.1 secondsSimulation length 60 seconds

strength, the same broadcasting period, the same message size fixed over time for all vehicles.Table 2 presents the values of the simulation parameters.

9

0

10

20

30

40

50

60

70

80

90

100

0 50 100

150 200

250 300

SM

R (

%)

VD i.e. #V in CR

CSMA/CA(SD)CSMA/CA(MD)

maxSMRminSMR

Figure 3: Successful Message Ratio (SMR) of the entire network with respect to the vehicledensity (VD) shows the communication reliability. maxSMR is the maximum possible SMRcalculated analytically. minSMR is the minimum possible SMR obtained by means of simulationsin which all vehicles have approximately the same phases for broadcasting. SD (Single Domain)case shows SMR degradation only due to the contention problem, where all the vehicles aredeployed at fixed locations within a single CR, therefore, there are no HNs. MD (Multi-Domain)shows SMR degradation due to both the contention and HN problems, where all vehicles aredeployed on a 3km long highway with three lanes per direction. Both cases show average valuesof ten simulations with a 99% confidence interval.

3.2 Simulation results and analysis

First, we study the reliability by means of the successful message reception ratio metric, i.e.,SMR([0,60)). The SMR of the overall network with respect to VD of the SD and MD cases areshown in Figure 3. For each different VD case, we performed ten simulations with a differentrandom seed for selecting the initial phases. Figure 3 presents the average values of thesesimulations with a confidence interval of 99%. In addition, the theoretical maximum SMR(maxSMR) is plotted to show the upper boundary. This maxSMR is given by

maxSMR(VD) =

{1 if VD ≤ SPSP/VD otherwise

, (21)

where SP is the channel saturation point, i.e., the maximum capacity of the channel in termsof the number of vehicles that can fit in one period duration without any overlap in time forbroadcasting. SP is given as

SP =T

Ts + Td, (22)

where Ts is the inter-frame space, i.e., an AIFS duration, and Td is the time to transmit asingle message. When all vehicles are optimally synchronized over the period for broadcasting,the SMR should approach this maxSMR level. Besides, we obtained the minimum possible SMRlevel (minSMR) by means of simulations in which we defined approximately the same phasesfor all vehicles.

From Figure 3, the HN problem appears to be the main cause of SMR degradation when the

10

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10 20

30 40

50 60

70 80

90 100

cdfS

MR

(60,S

MR

)

SMR (%)

CSMA/CA(MD)Ideal case(MD)

Figure 4: This shows the fairness through CDF of vehicles by their SMR, when VD is about 50.In the graph, a point indicates that y% of vehicles have at most x% SMR. The result shown is anaverage of ten simulations with a confidence interval of 99%. In an ideal fair case, the dashedline is expected where all vehicle should have the same SMR that is equal to SMR of the entirenetwork.

network is unsaturated. Once the network load exceeds its maximum capacity, the NN collisionsstart occurring in bursts thus yielding lower SMR.

We now continue our study at individual vehicle level to investigate the fairness. Here, weselect an unsaturated network condition where the traffic density is sparse, i.e., VD is about 50vehicles (that corresponds to about 85 vehicles per km over 6 lanes in our settings). Figure 4shows a relatively unfair distribution of message receptions over vehicles where some vehicleshave a high SMR whereas others have a relatively low SMR. In an ideal fair case, the dashedline is expected where the distance between the best and worst cases should be close to 0, i.e.,a transition from 0 to 1 within a short interval. In this case, however, the distance between thebest case and worst cases is approximately 65%.

Figure 5 shows how SMR of an individual vehicle evolves over a time interval of 15 seconds.We take only an interval of 15 seconds from a simulation of 60 seconds for better visibility.The graph shows two specific cases of vehicles with the best SMR and the worst SMR. Fromthis graph, we can conclude that vehicles remain in one condition (a certain SMR level) for arelatively long period of time if we ignore some fluctuations there. In addition, we can say thatthe best and the worst vehicles performances are clearly distinguished.

Figures 6 and 7 show the impact of the collision problems on the delay aspects at the linklevel through a cumulative distribution of links by their NoM and a histogram of links by theirFD respectively. During 60 seconds of simulation (VD=50, i.e., vehicles in total for a simulation),approximately 53000 links are established in total. Note, the link is an one-way relationship.Some vehicles join the CR of a vehicle whereas some may leave the CR due to the relativespeed between the vehicle and its neighbors. From Figure 6, we can see that almost 30% oflinks experience more than one second of NoM. This implies that a certain vehicle does notreceive a sequence of messages from another vehicle although the vehicle could have receivedthese messages in the absence of interferences. From Figure 7, many vehicles, i.e., about 350±50are seen that did not even discover some of their one-hop neighboring vehicles for their entire

11

0

10

20

30

40

50

60

70

80

90

100

15 20

25 30

SM

R (

%)

Timeline (seconds)

The best SMR(MD)The worst SMR(MD)

Figure 5: SMR fluctuations over time of two individual vehicles that experience the best and theworst SMR, respectively, when VD is approximately 50. It shows that there is a clear differencebetween the best and the worst cases. The graph shows the results of an arbitrary simulation.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8 9 10

CD

F o

f li

nks

NoM (seconds)

CSMA/CA(MD)

Figure 6: CDF of links by their NoM (the longest no message interval), when VD is approxi-mately 50. In the graph, a point indicates that y% of links have at most x seconds of NoM. Theresult shown is an average of ten simulations with a confidence interval of 99%.

encounter interval.

3.3 Summary of the evaluation

This section summarizes the main findings of the evaluation. We conclude the following:

• The HN problem is the main cause of the SMR degradation when the network is unsat-urated. Once it is saturated, the NN problem reduces the SMR dramatically. Therefore,the latter one is more a network congestion problem. In fact, this congestion problem iswell-known and addressed in many works, e.g., [5], [7], and [21]. The main approaches are

12

0 3000 6000 9000

12000 15000 18000 21000 24000 27000 30000 33000 36000

[0:0.1]

(0.1:0.2]

(0.2:1]

(1:5]

5< Never

Num

ber

of

links

Interval of FDs i.e. FirstDelay (seconds)

CSMA/CA(MD)

Figure 7: Distribution of links by their FD (delay to discover a new neighbor), when VD is ap-proximately 50. The graph presents 5 different intervals of FD. The last interval “Never” meansthat some vehicles never discover its neighbors. The number of links for “5<” and “Never” are1280±95 and 350±50, respectively. The result shown is an average of ten simulations with aconfidence interval of 99%.

to reduce beacon generation rate, beacon size, or to reduce the CR which are indeed allderived from (22).

• The impact of the HN problem is clearly revealed by an unfair SMR distribution and thedelay characteristics such as long lasting no message interval and first delay.

• The aforementioned quality impacts are mainly due to synchronized HNs, because vehiclestraveling on a highway, particulary those traveling in the same direction could have arather static topology for a relatively long period, i.e., in the order of multiple seconds2.In that topology, some vehicles could be incidentally synchronized as HNs which leads toa systematic message loss. It can be explained by a simple scenario where two vehiclesare moving forward in the same direction as illustrated in Figure 8. Assume further thatthe difference between the message broadcasting phases (φ) of the vehicles is less than amessage transmission time that they do not have any contenders (i.e., no vehicle in theCR with the same phase). In that scenario, both vehicles would remain broadcastingnearly at the same time and the messages would always collide at receiving vehicles in theintersection of the CRs as described in Figure 1c.

4 Solution study on the collision problem in oPBC

We are interested in the collision problems, particularly the HN problem, when the traffic densityis moderate or sparse, i.e., when the network is unsaturated. We assume that in that situationmessage loss is even more serious in terms of vehicle safety since the vehicles can have relatively

2When CR is 300m, two vehicles approaching each other from the opposite directions with a relative speed of80m/s will have an encounter interval of 7.5s.

13

- a synchronized HN

a

c d

e f b

- lane changing

Figure 8: This figure illustrates two examples of dangerous situations caused by synchronizedHNs. Vehicles “a” and “b” are synchronized HNs. In the picture, vehicle “d” is changing lanesassuming it is safe to do so. Because ”b is synchronized with ”a”, the driver of ”d” is notinformed about ”b”. Another case is a forward collision situation, where vehicle “a” is slowingdown but “e” and “f” are not aware of this.

high speeds. Therefore, such traffic conditions should have even stricter requirements on thecommunication. In the following sections, we look into three broadcasting schemes that canalleviate the impact of the collision problems. The key idea of these schemes is to break thesynchronization between vehicles as much as possible to prevent the systematic message loss.

4.1 Scheme-1: Elastic scheme

The first approach is what we call the elastic scheme in which the initial phase of broadcastingis changed at a regular basis. In this scheme, the message activation time is defined as follows

a(k)i

def=

φi if k = 0

a(k−1)i + Ti if k > 0 ∧ (k + φe) mod eri 6= 0

a(k−1)i + r(2Ti) if k > 0 ∧ k mod eri = 0

(23)

where eri is the elastic rate that defines how often the phase should be changed and r is afunction that returns a random value within the given interval. This value defines how much thephase should be changed. φe is a phase for starting elasticity and it is given as φe = br(eri)e.To keep the expected number of generated messages the same as the strict periodic scheme, 2Tiis selected as the interval. The worst case delay between two messages is 2Ti.

Figure 9, 11, and 12 show the results of this scheme in which we use the same er for allvehicles. From these graphs, we can make several interesting observations.

First, it is clearly seen that the more often the phase is changed, the better the elasticscheme improves the fairness and the delay characteristics. Particularly, the fairness is improveddrastically even at the higher value of er. The reason for this result is the frequent change ofphasing in the elastic scheme which affects the channel condition of the vehicle. Under thefrequent change of the channel condition, the lifetime of a synchronized period of the vehicles(also a period of favorable channel condition of the vehicle) becomes shorter, i.e., highly likely tobe at most the er period. Figures 10 and 11 reflect the effect of the short living synchronizationwhen er is 6; we see somewhat discrete and step-like effects. As a result, each vehicle experiencesmore or less the same fluctuating channel conditions in the long run.

14

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10 20

30 40

50 60

70 80

90 100

cdfS

MR

(60,S

MR

)

SMR (%)

CSMA/CA(MD)Ideal case(MD)

+Elastic(MD, er=2)+Elastic(MD, er=6)

Figure 9: The elastic scheme improves the fairness drastically, when VD is approximately 50.The result shown is an average of ten simulations with a confidence interval of 99%. For theelastic scheme, the confidence interval is ±0.3.

0

10

20

30

40

50

60

70

80

90

100

0 5 10 15

SM

R (

%)

Timeline (seconds)

The best SMR(MD, er=6)The worst SMR(MD, er=6)

Figure 10: SMR fluctuations over time of two individual vehicles that experience the best andthe worst SMR, respectively, when VD is approximately 50. Compared to the pure CSMA/CAcase, it is now hard to see any difference between the best and the worst cases. The graph showsthe results of an arbitrary simulation.

Second, in Figure 9 we can see that the elastic scheme does not affect SMR of the entirenetwork. It only affects SMRs of individual vehicles. For example, in the case of pure CSMA/CA,roughly half of the vehicles shows SMRs between 75-100%, while the other half shows SMRsbetween 40-75%. But, in the case of elastic scheme, this is completely changed and all vehiclesshow more or less the same SMRs that is closer to SMR of the entire network. This is alsoseen in Figure 10 where the difference between the best and the worst cases are now hardlydistinguishable.

15

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8

CD

F o

f li

nks

NoM (seconds)

CSMA/CA(MD)+Elastic(MD, er=6)+Elastic(MD, er=2)

Figure 11: The elastic scheme improves the NoM significantly, when VD is approximately 50.The result shown is an average of ten simulations with a confidence interval of 99%.

0 3000 6000 9000

12000 15000 18000 21000 24000 27000 30000 33000 36000

[0:0.1]

(0.1:0.2]

(0.2:1]

(1:5]

5< Never

Num

ber

of

links

Interval of FDs i.e. FirstDelay (seconds)

CSMA/CA(MD)+Elastic(MD, er=6)+Elastic(MD, er=2)

Figure 12: The elastic scheme improves FD significantly, when VD is approximately 50. In caseof “er=6”, the number of cases for “5<” and “Never” are 15±7 and 0, respectively. In case of“er=2”, the number of links for “5<” and “Never” are both 0. The result shown is an averageof ten simulations with a confidence interval of 99%.

4.2 Scheme-2: Jitter scheme

The second approach is what we call the jitter scheme in which the activation time is defined as

a(k)i = φi + kTi +AJi − r(2AJi), (24)

where AJi is an activation jitter that has a granularity of one message transmission time (i.e.,AJ = N ↔ AJ = NTd). The worst delays between messages of this scheme, therefore, is equalto Ti + 2AJi.

16

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10 20

30 40

50 60

70 80

90 100

cdfS

MR

(60,S

MR

)

SMR (%)

CSMA/CA(MD)Ideal case(MD)

+Jitter(MD, AJ=2)+Jitter(MD, AJ=20)

Figure 13: The jitter scheme improves the fairness, when VD is approximately 50. However, itshows that a small jitter does not help much for improving the fairness. The result shown is anaverage of ten simulations with 99% confidence interval. For the jitter scheme, the confidenceinterval is ±1.0.

Again, we can make a number of observations. First, similar as the elastic scheme, the jitterscheme improves the fairness and the delay characteristics as shown in Figure 13, 14, 15 and 16.We chose the same AJ for all vehicles. The bigger AJ is chosen, the better the jitter schemeworks. Note that a small jitter size does not show much improvement. Compared to the elasticscheme, the jitter scheme needs a bigger jitter size to improve the fairness though a small jittersize already works pretty well on the delay characteristics. This indeed makes sense, because,in the jitter scheme, the channel condition of a vehicle does not change completely comparedto the elastic scheme. Let’s say there are two vehicles synchronized with each other causingmessage collisions on their receivers. For the elastic scheme, we showed that the lifetime ofsuch synchronization becomes relatively short. But, in the jitter scheme, the two vehicles wouldremain synchronized during their entire encounter interval. The jitter only sometimes helps toprevent the message collisions happening. In addition, we can say that the jitter scheme worksbetter than the elastic scheme on the delay characteristics. Particularly, from Figure 16 we learnthat the number of links on a 0.2-1s interval is much lower than the elastic scheme result.

17

0

10

20

30

40

50

60

70

80

90

100

0 5 10 15

SM

R (

%)

Timeline (seconds)

The best SMR(MD, AJ=20)The worst SMR(MD, AJ=20)

Figure 14: SMR fluctuations over time of two individual vehicles that experience the best andthe worst SMR, respectively, when VD is approximately 50. Compared to the elastic scheme, itlooks more volatile. The graph shows the results of an arbitrary simulation.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8

CD

F o

f li

nks

NoM (seconds)

CSMA/CA(MD)+Jitter(MD, AJ=2)

+Jitter(MD, AJ=20)

Figure 15: The jitter scheme improves the NoM significantly, when VD is approximately 50.The jitter scheme performs better than the elastic scheme for this metric. In case of AJ=20,60% of the links have less than 0.5s of NoM. The result shown is an average of ten simulationswith a confidence interval of 99%.

4.3 Scheme-3: Elastic + Jitter (EJ) scheme

In addition to the previous two schemes, we also look into a third approach which is a combina-tion of the elastic and the jitter schemes. We call this scheme as the EJ scheme and it is defined

18

0 3000 6000 9000

12000 15000 18000 21000 24000 27000 30000 33000 36000

[0:0.1]

(0.1:0.2]

(0.2:1]

(1:5]

5< Never

Num

ber

of

links

Interval of FDs i.e. FirstDelay (seconds)

CSMA/CA(MD)+Jitter(MD, AJ=2)

+Jitter(MD, AJ=20)

Figure 16: The jitter scheme improves the FD significantly, when VD is approximately 50. Incase of AJ=2, the number of cases for “5<” and “Never” are 31±6 and 29±9, respectively. Incase of AJ=20, the number of cases for “5<” and “Never” are both 0. It shows that the jitterscheme performs better than the elastic scheme. In case of AJ=20, the number of the links inan interval of (0.2:1] is much lower. The result shown is an average of ten simulations with aconfidence interval of 99%.

as

a(k)i

def=

φi if k = 0

a(k−1)i + Ti +AJi − r(2AJi) if k > 0, (k + φe) mod eri 6= 0

a(k−1)i + r(2Ti) +AJi − r(2AJi) if k > 0, k mod eri = 0

(25)

As hoped, this solution outperforms both previous schemes as shown in Figure 17, 18, 20, and19. This third solution features the advantages of both schemes. Similar as the elastic scheme,it does improve the fairness drastically. Similar as the jitter scheme, it improves the delaycharacteristics to a greater extent.

.

19

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10 20

30 40

50 60

70 80

90 100

cdfS

MR

(60,S

MR

)

SMR (%)

Ideal case(MD)+EJ(MD, er=6, AJ=20)

+Elastic(MD,er=6)+Jitter(MD, AJ=20)

CSMA/CA(MD)

Figure 17: The EJ scheme outperforms the jitter scheme and it is slightly better than the elasticscheme for improving the fairness. The result shown is an average of ten simulations with aconfidence interval of 99%. For the elastic scheme, the confidence interval is ±0.3.

0

10

20

30

40

50

60

70

80

90

100

0 5 10 15

SM

R (

%)

Timeline (seconds)

The best SMR(MD, er=6, AJ=20)The worst SMR(MD, er=6, AJ=20)

Figure 18: SMR fluctuations over time of two individual vehicles that experience the best andthe worst SMR, respectively, when VD is approximately 50. The graph shows the results of anarbitrary simulation.

5 Related works

Performance studies of oPBC under CSMA/CA are carried out in many works. Here, we discussthe most relevant to our work. Computer simulation studies are done in [22, 17, 23] , analyticalmodels for analyzing performance of oPBC are proposed in [20, 19] and a real-world experimentis carried out in [4]. In [22], J. Yin et al., measured message throughput and latency under anunsaturated network condition. The throughput is defined as the long-run average percentage ofsingle-hop neighbors who successfully receive a broadcast message (SMR of the entire network).

20

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.5 1 1.5

2 2.5

CD

F o

f li

nks

NoM (seconds)

+Jitter(MD, AJ=20)+Elastic(MD, er=6)

+EJ(MD, er=6, AJ=20)

Figure 19: The EJ scheme improves the delay characteristic by reducing NoM similar as the jitterscheme, when VD is ap- proximately 50. The result shown is an average of ten simulations witha confidence interval of 99%.

0 3000 6000 9000

12000 15000 18000 21000 24000 27000 30000 33000 36000

[0:0.1]

(0.1:0.2]

(0.2:1]

(1:5]

5< Never

Num

ber

of

links

Interval of FDs i.e. FirstDelay (seconds)

+Jitter(MD, AJ=20)+Elastic(MD, er=6)

+EJ(MD, er=6, AJ=20)

Figure 20: The EJ scheme improves the delay characteristics similar as the jitter scheme. whenVD is ap- proximately 50. In case of EJ scheme, the number of links for “5<” and “Never” areboth 0, respectively. The result shown is an average of ten simulations with a confidence intervalof 99%.

The latency is defined as the long-run average time elapsing between sending a message at the

source and successfully receiving (f(k)i − s(k)i + δ) that message at a single-hop neighbor. They

conclude that the latency is reasonably good whereas the throughput is moderate and needsimprovement. It is not clear from their work what caused the moderate throughput. In [17], K.Bilstrup et al., analyzed a highway scenario with periodic broadcast of beacon messages. They

particularly studied the “channel access time” (s(k)i − a

(k)i ) under saturated network conditions

showing that a specific vehicle is forced to drop over 80% of its beacon messages because no

21

channel access was possible before the next message was generated. As a result, they confirmedthat some vehicles become invisible to surrounding vehicles for up to 10 seconds. Notably, theanalysis is done at vehicle level rather than entire network level. They did not, however, considerthe reception of messages. We think that “channel access time” is not a sufficient metric since thesuccessful channel access does not guarantee the successful message delivery due to the collisionand environmental problems. Also, it is more relevant to consider unsaturated conditions. In[23], T. Kuge et al., points out the impact of the HN problem on successful message receptionin oPBC under unsaturated network conditions. They used a metric that is defined as “the rateof the messages which are received correctly against all the messages that are generated duringmeasurement time (SMR of the entire network)”. They confirmed that the number of messagecollisions that are caused by the HN problem is higher than that are caused by the contentionproblem. Moreover, they showed that changing CW size helps to reduce the collisions causedby contention. But, it has no effect on the collisions caused by the HN problem. Instead, theyshowed that the collision rate can be reduced to up to one-third of the conventional CSMAscheme by using a spread spectrum (SS) scheme. They do not provide any analysis on delaycharacteristics.

Xiaomin Ma et al., propose an analytic model in [19] for analyzing performance of emergencymessage transmission as well as oPBC. In their model, they take the 802.11 backoff countingprocess, the HN problem, unsaturated network, fading channel, and mobility into account. Theyuse two metrics: a message reception rate that is defined as the ratio of the number of messagessuccessfully received by all vehicles within the range of a sending vehicle to the number ofmessages transmitted (SMR of the entire network), and a message transmission delay that is theaverage delay a message experiences between the time at which the message is generated and

the time at which the message is successfully received (f(k)i − a(k)i + δ). With their model, they

confirmed that the message delivery delay (less than 2ms) meets the requirement of the safetyapplications (500ms). However, the obtained message reception rates fail to meet reliabilityrequirements for the safety critical messaging. A. Vinel et al., propose an analytic model in [20]for analyzing performance of oPBC under both saturated and unsaturated network conditionsby two metrics: beacon message successful reception probability (SMR of the entire network)and mean beacon message transmission delay that is the time from the moment the beacon was

issued until it has been transmitted (f(k)i − a(k)i ). With the model, they found that the delay

requirements are met, but the probability of successful message reception is rather low in typicalscenarios. It is not clear whether they considered the HN problem in their model.

A very interesting real-world experiment is performed by F. Bai et al., in [4]. The workstudied the environmental impact (e.g., fading, doppler, multi-path effect) on reliability of the802.11p by an experimental set-up that includes a fleet of only three vehicles equipped withthe 802.11p communication system. With this setup, the probability of having the collisionproblems is almost 0. In that sense, our work is complementary to this work since we studythe impact of the collision problems (i.e., the other main reason of message loss besides theenvironmental impacts) on the quality of oPBC. The work judges the reliability of 802.11pusing distribution of consecutive message losses (similar to NoM. Note, NoM is the longest nomessage interval.) together with the general metric “message delivery ratio”. The messagedelivery ratio is calculated as a ratio of the number of data messages received at the receiverto total number of messages transmitted at the sender within some pre-defined time window(SMRij(I)). The work shows that the reliability of the 802.11p is adequate in a wide varietyof traffic environments. Most importantly, they observed that the message losses do not occursystematically viz., there are almost no consecutive message losses between two certain vehicleseven under the harsh freeway traffic environment.

22

All in all, we conclude the following: Two common metrics are widely used in these works.The first one is SMR of the entire network and the second one is an end-to-end delay (e.g., some

define it as f(k)i − s(k)i + δ and some define it as f

(k)i − a(k)i + δ or as f

(k)i − a(k)i ). These common

metrics do not show or quantify the exact causes of a certain problem (e.g., the contention andHN problems). Namely, SMR of the entire network cannot show the fairness aspect and theend-to-end delay metric cannot show no message interval between two vehicles which is morerelevant to vehicle safety applications that rely on oPBC. In addition, most of these works area lack of formality.

23

6 Concluding remarks

In this final section, we summarize the main findings and achievements of this work. In addition,we shed some light on future research directions.

6.1 Main contributions

We regard the following four results as the main contributions of the paper. Among these, thebiggest contribution of the work are the repairing schemes for improving the quality of oPBC.

First, a novel timing model of oPBC in the context of the 802.11p MAC has been defined.This model gives the foundation for a simulation model and for defining new more appropriatemetrics to judge the communication quality from the perspective of the safety applications.Among others, we introduce concepts of “neighborhood of a vehicle”, i.e., set of other vehiclesthat could receive its messages and “link”, i.e., an encounter interval that starts when a vehicleB enters the Communication Range (CR) of another vehicle A and ends when vehicle B leavesthe CR of vehicle A.

Second, new metrics based on the above concepts have been defined for judging the quality ofoPBC in three of the most important aspects: Successful Message Ratio (SMR) for the reliability,cumulative distribution of SMR per vehicle for the fairness, No Message interval (NoM) (i.e.,the longest interval in which no message is successfully delivered in a link), and First Delay (FD)(initial NoM) for the delay aspects respectively.

Third, an evaluation of oPBC is performed according to the simulation model under a strictperiodic broadcasting scheme, i.e., each vehicle broadcasts messages in a strictly periodic manner.The evaluation reveals that the HN problem is the main cause of various quality degradationsespecially when the network is unsaturated. Once the network is saturated, the contentionproblem already reduces the SMR dramatically. A detailed evaluation is conducted on anunsaturated network condition where the traffic condition is sparse, i.e., 85 vehicles/km ona highway with three lanes per direction. We selects such condition because we assume that itis a typical highway condition. Besides, this condition is even more stringent in terms of vehiclesafety since then vehicles have relatively high speeds. Such traffic condition, therefore, willhave even stricter requirements on the communication quality, particularly, on the delay aspect.The evaluation results suggest that the HN problem causes unfair SMR distribution where thedifference between the best and worst vehicles by their SMR is 65% in a simulation. Moreover,it causes long lasting consecutive message losses in a link of two vehicles. In some serious cases,a certain vehicle could not successfully deliver any of its messages to a particularly destinationvehicle throughout an entire link interval. These quality issues are mainly due to synchronizedHNs that can occur under the strict periodic scheme.

Fourth, we propose three simple but effective broadcasting schemes (i.e., elastic scheme, jitterscheme and a combination of these two schemes) to alleviate the impact of the HN problem.Though the three solutions do not affect the SMR (or reliability aspect) of the entire network,they do show significant improvements on the fairness and the delay aspects. Particularly, thecombined scheme features the advantages of other two schemes for improving the communicationquality in terms of the fairness and the delay aspects. These solutions are fully compatible withthe 802.11p, i.e., they are application-level solutions and can therefore be easily introduced inpractice.

6.2 Future directions

We are projecting the following future directions as our next steps:

24

• investigation of further broadcasting schemes (e.g., location and direction based schedul-ing) particularly for avoiding the collision problems;

• investigation of further combinations of different broadcasting schemes and their compat-ibility to each other (e.g., location based + jitter + predictive coding scheme);

• investigation of more adaptive schemes for the network congestion problem (e.g., dynamicmessage size or dynamic channel allocation i.e., switching between control and servicechannel);

• testing the proposed solutions with realistic traffic simulations and different traffic condi-tions (e.g., real traces of highway or urban roads);

• generalization of the solutions to domains (e.g., Mobile Ad-hoc Network or Wireless SensorNetworking).

7 Acknowledgement

We would like to thank Strategic Platform for Intelligent Traffic Systems (SPITS) project forfunding this work (spits-project.com). SPITS is a Dutch project, tasked with creating ITSconcepts that can improve mobility and safety.

References

[1] “Final report,” Vehicle Safety Communications Project, Tech. Rep. DOT HS 810 591, 2006.

[2] The CAR 2 CAR Communication Consortium, http://www.car-to-car.org/, 2011.

[3] Internet ITS consortium, http://www.internetits.org/, 2011.

[4] F. Bai and H. Krishnan, “Reliability Analysis of DSRC Wireless Communication for Vehi-cle Safety Applications,” Intelligent Transportation Systems Conference, 2006. ITSC ’06.IEEE, pp. 355–362, 2006.

[5] C. Robinson, D. Caveney, L. Caminiti, G. Baliga, K. Laberteaux, and P. Kumar, “Effi-cient Message Composition and Coding for Cooperative Vehicular Safety Applications,”Vehicular Technology, IEEE Transactions on, vol. 56, no. 6, pp. 3244–3255, Nov. 2007.

[6] J. Mittag, F. Schmidt-Eisenlohr, M. Killat, M. Torrent-Moreno, and H. Hartenstein, “MACLayer and Scalability Aspects of Vehicular Communication Networks,” VANET: VehicularApplications and Inter-Networking Technologies, pp. 219–269, 2010.

[7] M. Torrent-Moreno, J. Mittag, P. Santi, and H. Hartenstein, “Vehicle-to-Vehicle Communi-cation: Fair Transmit Power Control for Safety-Critical Information,” Vehicular Technology,IEEE Transactions on, vol. 58, no. 7, pp. 3684–3703, Sept. 2009.

[8] “IEEE Draft Standard for Information Technology - Telecommunications and informationexchange between systems - Local and metropolitan area networks - Specific requirements- Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) spec-ifications Amendment : Wireless Access in Vehicular Environments,” IEEE UnapprovedDraft Std P802.11p /D11.0, Mar. 2010.

25

[9] “IEEE Trial-Use Standard for Wireless Access in Vehicular Environments (WAVE) - Re-source Manager,” IEEE Std 1609.1, pp. 1–63, 2006.

[10] “IEEE Trial-Use Standard for Wireless Access in Vehicular Environments - Security Servicesfor Applications and Management Messages,” IEEE Std 1609.2, pp. 1–105, 2006.

[11] “IEEE Trial-Use Standard for Wireless Access in Vehicular Environments (WAVE) - Net-working Services,” IEEE Std 1609.3, pp. 1–87, 2007.

[12] “IEEE Trial-Use Standard for Wireless Access in Vehicular Environments (WAVE) - Multi-Channel Operation,” IEEE Std 1609.4, pp. 1–74, 2006.

[13] “Dedicated Short Range Communications (DSRC) Message Set Dictionary,” SAE StandardJ2735, 2009.

[14] “IEEE Standard for Information Technology- Telecommunications and Information Ex-change Between Systems-Local and Metropolitan Area Networks-Specific Requirements-Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Speci-fications,” IEEE Std 802.11-1997, pp. 1–445, 1997.

[15] D. R. Choffnes and F. E. Bustamante, “An integrated mobility and traffic model for vehic-ular wireless networks,” Proceedings of the 2nd ACM international workshop on Vehicularad hoc networks, pp. 69–78, 2005.

[16] R. Baumann, S. Heimlicher, and M. May, “Towards Realistic Mobility Models for VehicularAd-hoc Networks,” 2007 Mobile Networking for Vehicular Environments, pp. 73–78, May2007.

[17] K. Bilstrup, E. Uhlemann, E. Strom, and U. Bilstrup, “Evaluation of the IEEE 802.11pMAC Method for Vehicle-to-Vehicle Communication,” Vehicular Technology Conference,2008. VTC 2008-Fall. IEEE 68th, pp. 1–5, 2008.

[18] Q. Chen, F. Schmidt-Eisenlohr, D. Jiang, M. Torrent-Moreno, L. Delgrossi, and H. Harten-stein, “Overhaul of IEEE 802.11 modeling and simulation in NS-2,” Proceedings of the 10th

ACM Symposium on Modeling, analysis, and simulation of wireless and mobile systems,pp. 159–168, 2007.

[19] X. Ma and X. Chen, “Delay and Broadcast Reception Rates of Highway Safety Applicationsin Vehicular Ad Hoc Networks,” 2007 Mobile Networking for Vehicular Environments, pp.85–90, May 2007.

[20] A. Vinel, D. Staehle, and A. Turlikov, “Study of Beaconing for Car-to-Car Communicationin Vehicular Ad-Hoc Networks,” Communications Workshops, 2009. ICC Workshops 2009.IEEE International Conference on, pp. 1–5, 2009.

[21] L. Yang, J. Guo, and Y. Wu, “Channel Adaptive One Hop Broadcasting for VANETs,”Intelligent Transportation Systems, 2008. ITSC 2008. 11th International IEEE Conferenceon, pp. 369–374, 2008.

[22] J. Yin, T. ElBatt, G. Yeung, B. Ryu, S. Habermas, H. Krishnan, and T. Talty, “Perfor-mance evaluation of safety applications over DSRC vehicular ad hoc networks,” VANET’04: Proceedings of the 1st ACM international workshop on Vehicular ad hoc networks, pp.1–9, 2004.

26

[23] T. Kuge, K. Ohno, and M. Itami, “An analysis of performance degradation caused byhidden terminal and its improvement in Inter-Vehicle communication,” Intelligent TransportSystems Telecommunications,(ITST),2009 9th International Conference on, pp. 482–485,2009.

Appendices

A CSMA/CA

This section explains how CSMA/CA operates in periodic broadcast mode. Figure 21 showsa basic flowchart and Figure 22 gives a corresponding state machine diagram. According toCSMA/CA, when a station becomes ready to broadcast (Ready To Transmit (RTT)), the stationmust first check the channel for a duration of AIFS. If the channel has been idle for longer thanAIFS, the station starts its transmission immediately. If the channel is busy or becomes busyduring AIFS, the station must wait for the channel to become idle. 802.11 refers to this wait asAccess Deferral. If access is deferred, the station first waits for the channel to become idle forAIFS again. If the channel is idle, the station must perform Bf procedure by starting a Bf timerwhich is set to a random number drawn from an interval of {0,1,. . . ,CW}. The timer has thegranularity of a slot time and is decremented every time when the channel is sensed to be idlefor a slot time. The timer is stopped in case the channel becomes busy and the decrementingprocess is resumed when the channel becomes idle again (i.e., idle for a duration of AIFS). Thestation is allowed to transmit its message when the Bf timer reaches zero. Depending on thechannel condition, a station may experience multiple AIFS plus access deferrals. Note that theBf counting is at most done once. If a new message arrives from the upper layer, then thecurrent message must be dropped and the new message transmission will start.

27

a=0?

Transmit

No

Yes

Yes

No

Listen CH for a slotTime;

c--;

Yes

c>0?

No

Yes

a:=AIFS;c:=0;do_c:=0;

(Ready To Transmit)

Listen CH for a slotTime;

a--;

c:={0,..,CW}

Drop the current

message

Defer the access;

do_c := 1; a:=AIFS;

Next msg from app?

Next msg from app?

No

No

Yes

Next msg from app?

No

CH idle?

CH idle?

No

Yes

Yes

do_c > 0 c=0?

Start

Figure 21: CSMA/CA procedure in periodic broadcast mode

28

DEFERRINGWAIT_AIFS

IDLE

Bf

COUNTING

XMIT

a=0 |

c=0do_c>0→c:={0,..,CW}

c=0 |

n+:=msgDelay

¬busyc>0 |

n+:= ST,

c--,

do_c:=0

busy

n:=NMT

RTT |

a:=AIFS,

c:=0,

do_c:=0

busy

n+:= busyTime,

do_c:=1

a:=AIFS

¬busya>0 |

n+:= ST,

a--

n+busyTime≥NMT |

n:=NMT

a>0n+ST≥NMT |

n:=NMT

c>0n+ST≥NMT

(c=0n+msgDelay>NMT) |

n:=NMT

RTT – Ready To Transmit

ST – a Slot Time

NMT – Next Message Time

Figure 22: CSMA/CA state machine

As shown in the state machine diagram in Figure 22, each station is in one of the five states:IDLE, WAIT AIFS, DEFERRING, Bf COUNTING, and XMIT. The station is in the IDLEstate when it is neither in transmission nor RTT. In this diagram, we use several variables andconstants to show the state transition conditions and timing changes.

• The “n” holds the current time.

• The “a” holds the current value of the AIFS counting.

• The “c” holds the current value of the Bf counting.

• The “do c” is a boolean variable to indicate whether the station should perform the Bfcounting.

• The “busyTime” is the duration of the channel being busy.

• The “msgDelay” is the duration of one message transmission.

• The “Next Message Time (NMT)” is the time at which the station becomes ready totransmit its next message.

• The “busy” to indicate the busy channel.

• The “AIFS” is the number of slots for the AIFS waiting and it is given by the standard.

29

• The “CW” is the number of slots for the Bf and it is given by the standard.

• The “a Slot Time (ST)” is the duration of one slot time and it is given by the standard.

B Signal reception model

There are generally two methods used for physical layer (or PHY) modeling in network simu-lations, namely SINR threshold based and Bit Error Rate (BER) based [22]. Under the formermethod, the receiver accepts the message when the computed SINR value is above the SINRthreshold for a particular modulation scheme. The method based on BER decides whether ornot a message is received successfully based on the message length and bit error rate deducedby the pre-computed BER versus SINR curve for every modulation scheme at the receiver.

In our case, the signal reception model is taken from the NS-2 implementation of 802.11p[18], where the signal reception decision is based on SINR ratio. In this model, three basic signalthreshold concepts play a role. The first one is a Power Sense Threshold, PsTh.3 If any receivingsignal is equal to or greater than PsTh, the PHY will try to decode the signal. In addition, anysignal equal to or greater than PsTh is considered to be strong enough to interfere with anyother signal. Therefore, such signal is added into the cumulative signal level of the receiver. Thesecond threshold is the SINR threshold (or receiving threshold), SrTh. To receive a messagesuccessfully, the preamble must be received successfully. While a station is not transmitting norreceiving any signal, namely the PHY is constantly searching for a preamble and if a new signalarrives with a signal strength that is equal to or greater than SrTh with respect to all otherinterfering signals plus noise, then the station will start receiving the signal as the preamble. Ifthis SINR ratio holds for the entire duration of the preamble length, the PHY can finish thepreamble reception successfully. Once the preamble is received successfully, the PHY will informthe MAC layer that the PHY is receiving a message and will continue until the complete payloadhas been received regardless of whether the signal level is greater or less than SrTh, unless theframe capturing feature is enabled. Note, the preamble includes information about payload (thelength of payload, modulation scheme etc). In addition, if the receiving signal strength becomesless than SrTh during the payload reception, PHY layer puts an error in the payload such thatwhen the MAC layer checks the CRC of the received message it will not send the message tothe application layer.

The third threshold is a carrier sense threshold, CsTh. If the cumulative power level of thereceiver is equal to or greater than CsTh, then the PHY layer will inform the MAC layer thatthe channel is busy. From this, we can see that the MAC layer is informed about the channelstatus in two different ways. First, no matter what if the cumulative power is equal to or greaterthan CsTh, then the MAC is informed the channel is busy. Second, if the PHY layer successfullyreceives a preamble viz., start receiving a payload, the MAC layer is informed the channel isbusy.

3Some concepts are rather misleading. In the document provided there are a RX threshold and a carrier sensethreshold. But in the implementation there are a SINR threshold, a power sense threshold and a carrier sensethreshold.

30